A review by researchers at Southern Illinois University Edwardsville (SIUE) have found that two recent studies examining purported cropland expansion in the Midwest are based on a flawed methodology that “suffers from accuracy and certainty issues.” The SIUE authors found that the inherent defects in the methodology of studies by Zhang et al. and Lark et al., “severely hinder its use for estimating land use change over time.”
In their paper, Joshua Pritsolas and Randall Pearson of SIUE’s GeoSpatial Mapping, Applications, and Research Center pointed out that both studies relied on the U.S. Department of Agriculture’s Cropland Data Layer (CDL) tool to estimate the conversion of grassland to cropland, a use for which the tool was not intended and is poorly suited. As the USDA itself has noted, “Unfortunately, the pasture and grass-related land cover categories have traditionally had very low classification accuracy in the CDL,” meaning grassland is often confused with cropland in the CDL dataset.
The reliance of Zhang et al. and Lark et al. on USDA’s CDL tool renders the results of both studies highly questionable. “Given these issues, policy makers should exercise caution in referencing studies that have performed or integrated land cover/use change analysis that relies on the CDL,” according to Pritsolas and Pearson.
According to the SIUE analysis, it is likely that Zhang et al. and Lark et al. grossly overstated the amount of cropland expansion between 2008 and 2016 because the CDL tool frequently misclassified cropland as grassland in the early part of this time period. “The cropland expansion claimed by Lark et al. (2020) and adopted by Zhang et al. (2021) has a high potential of being false change due to poor classification certainty in the earlier CDL,” the authors found.